When connected via SSH to my headless Ubuntu 20.04 machine with two Nvidia 3070 Ti cards, nvidia-smi shows
Fri Oct 29 00:14:25 2021
+---------------------------------------------+
| NVIDIA-SMI 460.91.03 Driver Version: 460.91.03 CUDA Version: 11.2 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|===============================+======================+======================|
| 0 GeForce RTX 307... Off | 00000000:09:00.0 Off | N/A |
| 31% 61C P2 171W / 174W | 6843MiB / 7982MiB | 100% Default |
| | | N/A |
+-------------------------------+----------------------+----------------------+
| 1 GeForce RTX 307... Off | 00000000:43:00.0 Off | N/A |
| 64% 80C P2 173W / 174W | 6853MiB / 7981MiB | 100% Default |
| | | N/A |
+-------------------------------+----------------------+----------------------+
The output of nvidia-smi -i 0 -q shows the core and memory clock speeds:
Clocks
Graphics : 1800 MHz
SM : 1800 MHz
Memory : 9251 MHz
Video : 1575 MHz
However, when I try to increase the memory clock speed of the first GPU by 1000 MHz using the command
nvidia-settings -c :0 -a '[gpu:0]/GPUMemoryTransferRateOffset[2]=1000'
we get the error:
Unable to init server: Could not connect: Connection refused
No protocol specified
ERROR: Unable to find display on any available system
No protocol specified
ERROR: Unable to find display on any available system
What is the proper way to change the memory clock speeds from the terminal?
Note:
- There are no monitors/keyboard/mouse connected to this machine
- There is no X server or desktop environment installed. If it is necessary to have them, the solution should still work via SSH/VNC and using the bash shell/terminal.
